3wolfMondAI-SDG
The reunion of StableMondAI-SDG and RealMondAI-SDG. This model supports toony and realism styles well, and comes with a pleasant default inbuilt style while being nearly as malleable as StableMondAI-SDG on artist and style tags.
Use the appropriate realism tags to "turn it on", or use artist tags without them to generate toony styles.
Disclaimer
I am not the creator of this model but are allowed to host it here on civitai.
If you have any questions, suggestions or want to leave some praise about this model head over here
Prompt Recommendations
Start with a basic prompt and work from there. This model works better with leaner prompts.
Prompts need to use danbooru/e621 tags, natural language prompts are not supported. Use noobAI-XL quality tags at your own discretion, they have some style influence.
Starter Positive prompt:
masterpiece, best quality, newest,Starter Negative prompt:
worst quality, blurry background,To enable realism mode add the following tags to the positive prompt:
(photorealism, realistic,)If you want increased fur detail especially in realism mode, try adding some of these tags into the positive prompt:
(detailed fur:1.2), fluffy, shaggy fur,For better generation of anthro characters it can be beneficial to add tags that describe human features into the negative prompt, e.g.
human teeth, human handsGeneration Recommendations
The following settings are recommended starting points:
Sampler: Euler A
Scheduler: SGM Uniform
Steps: 28-32
CFG: 5.5
RescaleCFG 0.6-0.7 is recommended.
Initial tests show that ancestral samplers work best but DPM can work with the right scheduler.
More Information
Check the HF for sampler comparisons and style grids.
Description
Initial version
FAQ
Comments (21)
You may notice that some species sometimes generate with canine/dog noses, this is a bias that is present even in base noobaiXL but seems to be exacerbated by the added realism training data. If you encounter this issue, put "canine nose" into the negative prompt, that usually gets rid of the problem. If you want to generate mixed duos you can still put it in because for any canine species the bias will overwrite the negative prompt.
Insanely good AI not only for anthro, but for demon/oni type characters. This is the model I've been looking for. Also, it absolutely NAILS the hands most of the time, which is rare. Good job!
This is a very interesting checkpoint, thanks for making it!
It is one of the best models out there. Capable of creating a wide range of style / characters / details.
Perfect for both Anthro and Human characters.
Great work !
Does this model work with A1111? When I attempt to duplicate a typical generation, I just end up with a colorful blob of pixels and not much of an image. I am thinking (and this is me as a nooby) that I am missing some additional tweak. thanks for any advice all!
I think A111 does not autodetect VPRED models, you will have to enable Zero SNR somewhere in the options for it to work. Maybe look into switching to Forge/reForge.
At first i didnt expect to like this model as much, since it mentioned being more focused on realism and i didnt like the realmondai model as much because of that... BUT luckily for me, i very much like this ones natural style for some things, with or without the realism tags, and it still can do all artist styles i tried very nicely.
Seems to have most v-pred capabilities, like stablemondai it seems more then some other noob v-pred finetunes. I didnt try anything extreme with lighting yet, but any level of day and night seem to work fine without much effort, i even got some very nice sunsets and things, and just pure black or white backgrounds work too. For me it has juuust the right amount of Illustrious influence to make it well, stable, but not enough to introduce many of its biases. Backgrounds also were great for me so far, which i think worked a bit less well for instance in stablemondai for me.
Anatomy seems very stable to me, even for some more niche species of furries, scalies and also ferals. This seems to have much less human bias then even some other noob finetunes, but it still wants to make things at least partly human, adding skin or facial features, if you put enough weight on the realism tags, but even then it still holds up great compared.
Small issue, when generating kobolds about 50% of the time the background will be a black and white sketch or empty. You'll need heavy prompting to help with it currently.
post an example. if I get sketches I usually just put greyscale,in the neg prompt and that solves it
Love the model, I highly encourage people to use this model.
It's crazy how jack-of-all-trades this checkpoint is. It's practically all I use.
Is any update planned? It's still the only checkpoint I never delete, top quality.
By the way, I had some issues with mangled hands and feet, bad nails, it was REALLY bad. And I noticed that this model is extremely prompt sensitive and the sampler you use matters. I got the best results (for a semirealistic style) using these settings:
Positive prompt start: masterpiece, very aesthetic, best quality, newest, absurdres, highres, very awa, (realistic, photorealistic)
Negative prompt start: lowres, worst quality, low quality, bad anatomy, bad hands, multiple views, abstract, signature, 2koma, 4koma, comic, manga, text, watermark, username, artist signature, patreon logo, patreon username, wrinkles, twitter username, buttons, green theme
25 steps
1.1 cfg
sampler: euler_ancestral_cfg_pp
scheduler: sgm_uniform
These are the best settings I could find after using this model regularly for so long. Follow these guidelines and you'll have some really clean results. You can get more realistic results with different prompts, but I never managed to get a realistic rendering with good hands/feet, even though the faces look noticeably better.
No new base models/finetunes have been released, about all the "new" models that appear on civit every other day are just re-merges upon re-merges of existing stuff in the search for pleasant built-in styles, but 99% of them still only share the common noobai-XL knowledge, nothing new (concepts/chars/artists) is being added. As long as there is no noobai-XL 2.0, there is not really any (easy/cheap) way to update it with more knowledge, its more flexible to just add a lora or two if you want to do something that is unknown to noobai-XL.
@Xeno443 Thank you for the reply! I'll definitely try to use some stylistic LORAs alongside it, never really did that before.
Getting canine spades with nostrils. Is this a model or software issue?
It is a known problem, cause the models will interpret the shape of the spade as a mouth, and as such want to add a nose. try lower denoise upscale/inpaint, or blob it over with an image editor before that.
@Xeno443 Ooh, thats why horizontal cloacas sometimes get... teeth... c: Not a big problem tho, still prefer this model over like... any other ive tried since its release.
It is a problem stemming even from base noobai, but in more toony-style oriented models its a bit less pronounced, the realism increases the chance of it happening, for whatever machine spirit related reason
@Xeno443 i was wondering what was causing it lol. does last sdg have the same issue?
@phRipper
It is usually not a problem on the rawgens, but if you upscale, inpaint or use adetailer, the model often misinterprets the general shape as being a mouth, and as such wants to add a nose. it is just trained much more on faces, so it is biased to recognized a lot of similar shapes as such. as it likes to misinterpret beans for other body parts too.
@Xeno443 ah, well I dont know how to inpaint yet.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.